X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmgs%2Fmgs_llog.c;h=726262f80bd8f44fca7629113d922e4ccdc76f51;hp=15102e057a2bbb57760ba3df007a712acf44a2f8;hb=9db2fa9133b15a99a1d5a04914964a015de32367;hpb=4c90aef2f0712d8da720f6a66cd09b88df7d0573 diff --git a/lustre/mgs/mgs_llog.c b/lustre/mgs/mgs_llog.c index 15102e0..726262f 100644 --- a/lustre/mgs/mgs_llog.c +++ b/lustre/mgs/mgs_llog.c @@ -23,7 +23,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2016, Intel Corporation. + * Copyright (c) 2011, 2017, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -42,10 +42,11 @@ #define D_MGS D_CONFIG #include -#include -#include +#include +#include #include #include +#include #include "mgs_internal.h" @@ -70,6 +71,7 @@ int class_dentry_readdir(const struct lu_env *env, struct mgs_device *mgs, struct mgs_direntry *de; char *key; int rc, key_sz; + size_t suffix_len = sizeof(".bak") - 1; INIT_LIST_HEAD(log_list); @@ -106,9 +108,8 @@ int class_dentry_readdir(const struct lu_env *env, struct mgs_device *mgs, } /* filter out ".bak" files */ - /* sizeof(".bak") - 1 == 3 */ - if (key_sz >= 3 && - !memcmp(".bak", key + key_sz - 3, 3)) { + if (key_sz >= suffix_len && + !memcmp(".bak", key + key_sz - suffix_len, suffix_len)) { CDEBUG(D_MGS, "Skipping backup file %.*s\n", key_sz, key); goto next; @@ -190,7 +191,7 @@ static int mgs_fsdb_handler(const struct lu_env *env, struct llog_handle *llh, int cfg_len = rec->lrh_len; char *cfg_buf = (char*) (rec + 1); struct lustre_cfg *lcfg; - __u32 index; + u32 index; int rc = 0; ENTRY; @@ -210,12 +211,14 @@ static int mgs_fsdb_handler(const struct lu_env *env, struct llog_handle *llh, CDEBUG(D_INFO, "cmd %x %s %s\n", lcfg->lcfg_command, lustre_cfg_string(lcfg, 0), lustre_cfg_string(lcfg, 1)); - /* Figure out ost indicies */ - /* lov_modify_tgts add 0:lov1 1:ost1_UUID 2(index):0 3(gen):1 */ - if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD || - lcfg->lcfg_command == LCFG_LOV_DEL_OBD) { - index = simple_strtoul(lustre_cfg_string(lcfg, 2), - NULL, 10); + /* Figure out ost indicies */ + /* lov_modify_tgts add 0:lov1 1:ost1_UUID 2(index):0 3(gen):1 */ + if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD || + lcfg->lcfg_command == LCFG_LOV_DEL_OBD) { + rc = kstrtouint(lustre_cfg_string(lcfg, 2), 10, &index); + if (rc) + RETURN(rc); + CDEBUG(D_MGS, "OST index for %s is %u (%s)\n", lustre_cfg_string(lcfg, 1), index, lustre_cfg_string(lcfg, 2)); @@ -541,14 +544,15 @@ int mgs_cleanup_fsdb_list(struct mgs_device *mgs) return 0; } -int mgs_find_or_make_fsdb(const struct lu_env *env, struct mgs_device *mgs, - char *name, struct fs_db **dbh) +/* The caller must hold mgs->mgs_mutex. */ +int mgs_find_or_make_fsdb_nolock(const struct lu_env *env, + struct mgs_device *mgs, + char *name, struct fs_db **dbh) { struct fs_db *fsdb; int rc = 0; ENTRY; - mutex_lock(&mgs->mgs_mutex); fsdb = mgs_find_fsdb(mgs, name); if (!fsdb) { fsdb = mgs_new_fsdb(env, mgs, name); @@ -557,7 +561,6 @@ int mgs_find_or_make_fsdb(const struct lu_env *env, struct mgs_device *mgs, CDEBUG(D_MGS, "Created new db: rc = %d\n", rc); } - mutex_unlock(&mgs->mgs_mutex); if (!rc) *dbh = fsdb; @@ -565,6 +568,19 @@ int mgs_find_or_make_fsdb(const struct lu_env *env, struct mgs_device *mgs, RETURN(rc); } +int mgs_find_or_make_fsdb(const struct lu_env *env, struct mgs_device *mgs, + char *name, struct fs_db **dbh) +{ + int rc; + ENTRY; + + mutex_lock(&mgs->mgs_mutex); + rc = mgs_find_or_make_fsdb_nolock(env, mgs, name, dbh); + mutex_unlock(&mgs->mgs_mutex); + + RETURN(rc); +} + /* 1 = index in use 0 = index unused -1= empty client log */ @@ -616,6 +632,28 @@ static __inline__ int next_index(void *index_map, int map_len) return -1; } +/* Make the mdt/ost server obd name based on the filesystem name */ +static bool server_make_name(u32 flags, u16 index, const char *fs, + char *name_buf, size_t name_buf_size) +{ + bool invalid_flag = false; + + if (flags & (LDD_F_SV_TYPE_MDT | LDD_F_SV_TYPE_OST)) { + if (!(flags & LDD_F_SV_ALL)) + snprintf(name_buf, name_buf_size, "%.8s%c%s%04x", fs, + (flags & LDD_F_VIRGIN) ? ':' : + ((flags & LDD_F_WRITECONF) ? '=' : '-'), + (flags & LDD_F_SV_TYPE_MDT) ? "MDT" : "OST", + index); + } else if (flags & LDD_F_SV_TYPE_MGS) { + snprintf(name_buf, name_buf_size, "MGS"); + } else { + CERROR("unknown server type %#x\n", flags); + invalid_flag = true; + } + return invalid_flag; +} + /* Return codes: 0 newly marked as in use <0 err @@ -684,7 +722,7 @@ static int mgs_set_index(const struct lu_env *env, clear_bit(FSDB_LOG_EMPTY, &fsdb->fsdb_flags); if (server_make_name(mti->mti_flags & ~(LDD_F_VIRGIN | LDD_F_WRITECONF), mti->mti_stripe_index, mti->mti_fsname, - mti->mti_svname)) { + mti->mti_svname, sizeof(mti->mti_svname))) { CERROR("unknown server type %#x\n", mti->mti_flags); GOTO(out_up, rc = -EINVAL); } @@ -913,7 +951,7 @@ static int mgs_modify(const struct lu_env *env, struct mgs_device *mgs, GOTO(out_free, rc = -E2BIG); /* Modify mostly means cancel */ mml->mml_marker.cm_flags = flags; - mml->mml_marker.cm_canceltime = flags ? cfs_time_current_sec() : 0; + mml->mml_marker.cm_canceltime = flags ? ktime_get_real_seconds() : 0; mml->mml_modified = 0; rc = llog_process(env, loghandle, mgs_modify_handler, (void *)mml, NULL); @@ -933,18 +971,23 @@ out_pop: RETURN(rc); } +enum replace_state { + REPLACE_COPY = 0, + REPLACE_SKIP, + REPLACE_DONE, + REPLACE_UUID, + REPLACE_SETUP +}; + /** This structure is passed to mgs_replace_handler */ -struct mgs_replace_uuid_lookup { +struct mgs_replace_data { /* Nids are replaced for this target device */ struct mgs_target_info target; /* Temporary modified llog */ struct llog_handle *temp_llh; - /* Flag is set if in target block*/ - int in_target_device; - /* Nids already added. Just skip (multiple nids) */ - int device_nids_added; - /* Flag is set if this block should not be copied */ - int skip_it; + enum replace_state state; + char *failover; + char *nodeuuid; }; /** @@ -952,42 +995,42 @@ struct mgs_replace_uuid_lookup { * b) is it target block * * \param[in] lcfg - * \param[in] mrul + * \param[in] mrd * * \retval 0 should not to be skipped * \retval 1 should to be skipped */ static int check_markers(struct lustre_cfg *lcfg, - struct mgs_replace_uuid_lookup *mrul) + struct mgs_replace_data *mrd) { struct cfg_marker *marker; /* Track markers. Find given device */ if (lcfg->lcfg_command == LCFG_MARKER) { marker = lustre_cfg_buf(lcfg, 1); - /* Clean llog from records marked as CM_EXCLUDE. - CM_SKIP records are used for "active" command + /* Clean llog from records marked as CM_SKIP. + CM_EXCLUDE records are used for "active" command and can be restored if needed */ - if ((marker->cm_flags & (CM_EXCLUDE | CM_START)) == - (CM_EXCLUDE | CM_START)) { - mrul->skip_it = 1; + if ((marker->cm_flags & (CM_SKIP | CM_START)) == + (CM_SKIP | CM_START)) { + mrd->state = REPLACE_SKIP; return 1; } - if ((marker->cm_flags & (CM_EXCLUDE | CM_END)) == - (CM_EXCLUDE | CM_END)) { - mrul->skip_it = 0; + if ((marker->cm_flags & (CM_SKIP | CM_END)) == + (CM_SKIP | CM_END)) { + mrd->state = REPLACE_COPY; return 1; } - if (strcmp(mrul->target.mti_svname, marker->cm_tgtname) == 0) { + if (strcmp(mrd->target.mti_svname, marker->cm_tgtname) == 0) { LASSERT(!(marker->cm_flags & CM_START) || !(marker->cm_flags & CM_END)); if (marker->cm_flags & CM_START) { - mrul->in_target_device = 1; - mrul->device_nids_added = 0; + mrd->state = REPLACE_UUID; + mrd->failover = NULL; } else if (marker->cm_flags & CM_END) - mrul->in_target_device = 0; + mrd->state = REPLACE_COPY; } } @@ -1068,57 +1111,117 @@ static inline int record_setup(const struct lu_env *env, * \retval 0 record is not processed. */ static int process_command(const struct lu_env *env, struct lustre_cfg *lcfg, - struct mgs_replace_uuid_lookup *mrul) + struct mgs_replace_data *mrd) { int nids_added = 0; lnet_nid_t nid; char *ptr; - int rc; + int rc = 0; - if (lcfg->lcfg_command == LCFG_ADD_UUID) { + if (mrd->state == REPLACE_UUID && + lcfg->lcfg_command == LCFG_ADD_UUID) { /* LCFG_ADD_UUID command found. Let's skip original command and add passed nids */ - ptr = mrul->target.mti_params; + ptr = mrd->target.mti_params; while (class_parse_nid(ptr, &nid, &ptr) == 0) { + if (!mrd->nodeuuid) { + rc = name_create(&mrd->nodeuuid, + libcfs_nid2str(nid), ""); + if (rc) { + CERROR("Can't create uuid for " + "nid %s, device %s\n", + libcfs_nid2str(nid), + mrd->target.mti_svname); + return rc; + } + } CDEBUG(D_MGS, "add nid %s with uuid %s, " "device %s\n", libcfs_nid2str(nid), - mrul->target.mti_params, - mrul->target.mti_svname); + mrd->target.mti_params, + mrd->nodeuuid); rc = record_add_uuid(env, - mrul->temp_llh, nid, - mrul->target.mti_params); + mrd->temp_llh, nid, + mrd->nodeuuid); if (!rc) nids_added++; + + if (*ptr == ':') { + mrd->failover = ptr; + break; + } } if (nids_added == 0) { CERROR("No new nids were added, nid %s with uuid %s, " "device %s\n", libcfs_nid2str(nid), - mrul->target.mti_params, - mrul->target.mti_svname); - RETURN(-ENXIO); + mrd->nodeuuid ? mrd->nodeuuid : "NULL", + mrd->target.mti_svname); + name_destroy(&mrd->nodeuuid); + return -ENXIO; } else { - mrul->device_nids_added = 1; + mrd->state = REPLACE_SETUP; } return nids_added; } - if (mrul->device_nids_added && lcfg->lcfg_command == LCFG_SETUP) { + if (mrd->state == REPLACE_SETUP && lcfg->lcfg_command == LCFG_SETUP) { /* LCFG_SETUP command found. UUID should be changed */ rc = record_setup(env, - mrul->temp_llh, + mrd->temp_llh, /* devname the same */ lustre_cfg_string(lcfg, 0), /* s1 is not changed */ lustre_cfg_string(lcfg, 1), - /* new uuid should be - the full nidlist */ - mrul->target.mti_params, + mrd->nodeuuid, /* s3 is not changed */ lustre_cfg_string(lcfg, 3), /* s4 is not changed */ lustre_cfg_string(lcfg, 4)); + + name_destroy(&mrd->nodeuuid); + if (rc) + return rc; + + if (mrd->failover) { + ptr = mrd->failover; + while (class_parse_nid(ptr, &nid, &ptr) == 0) { + if (mrd->nodeuuid == NULL) { + rc = name_create(&mrd->nodeuuid, + libcfs_nid2str(nid), + ""); + if (rc) + return rc; + } + + CDEBUG(D_MGS, "add nid %s for failover %s\n", + libcfs_nid2str(nid), mrd->nodeuuid); + rc = record_add_uuid(env, mrd->temp_llh, nid, + mrd->nodeuuid); + if (rc) { + name_destroy(&mrd->nodeuuid); + return rc; + } + if (*ptr == ':') { + rc = record_add_conn(env, + mrd->temp_llh, + lustre_cfg_string(lcfg, 0), + mrd->nodeuuid); + name_destroy(&mrd->nodeuuid); + if (rc) + return rc; + } + } + if (mrd->nodeuuid) { + rc = record_add_conn(env, mrd->temp_llh, + lustre_cfg_string(lcfg, 0), + mrd->nodeuuid); + name_destroy(&mrd->nodeuuid); + if (rc) + return rc; + } + } + mrd->state = REPLACE_DONE; return rc ? rc : 1; } @@ -1132,22 +1235,22 @@ static int process_command(const struct lu_env *env, struct lustre_cfg *lcfg, * * \param[in] llh log to be processed * \param[in] rec current record - * \param[in] data mgs_replace_uuid_lookup structure + * \param[in] data mgs_replace_data structure * * \retval 0 success */ -static int mgs_replace_handler(const struct lu_env *env, - struct llog_handle *llh, - struct llog_rec_hdr *rec, - void *data) +static int mgs_replace_nids_handler(const struct lu_env *env, + struct llog_handle *llh, + struct llog_rec_hdr *rec, + void *data) { - struct mgs_replace_uuid_lookup *mrul; + struct mgs_replace_data *mrd; struct lustre_cfg *lcfg = REC_DATA(rec); int cfg_len = REC_DATA_LEN(rec); int rc; ENTRY; - mrul = (struct mgs_replace_uuid_lookup *)data; + mrd = (struct mgs_replace_data *)data; if (rec->lrh_type != OBD_CFG_REC) { CERROR("unhandled lrh_type: %#x, cmd %x %s %s\n", @@ -1163,23 +1266,27 @@ static int mgs_replace_handler(const struct lu_env *env, GOTO(skip_out, rc = 0); } - rc = check_markers(lcfg, mrul); - if (rc || mrul->skip_it) + rc = check_markers(lcfg, mrd); + if (rc || mrd->state == REPLACE_SKIP) GOTO(skip_out, rc = 0); /* Write to new log all commands outside target device block */ - if (!mrul->in_target_device) + if (mrd->state == REPLACE_COPY) GOTO(copy_out, rc = 0); - /* Skip all other LCFG_ADD_UUID and LCFG_ADD_CONN records - (failover nids) for this target, assuming that if then - primary is changing then so is the failover */ - if (mrul->device_nids_added && + if (mrd->state == REPLACE_DONE && (lcfg->lcfg_command == LCFG_ADD_UUID || - lcfg->lcfg_command == LCFG_ADD_CONN)) + lcfg->lcfg_command == LCFG_ADD_CONN)) { + if (!mrd->failover) + CWARN("Previous failover is deleted, but new one is " + "not set. This means you configure system " + "without failover or passed wrong replace_nids " + "command parameters. Device %s, passed nids %s\n", + mrd->target.mti_svname, mrd->target.mti_params); GOTO(skip_out, rc = 0); + } - rc = process_command(env, lcfg, mrul); + rc = process_command(env, lcfg, mrd); if (rc < 0) RETURN(rc); @@ -1187,7 +1294,7 @@ static int mgs_replace_handler(const struct lu_env *env, RETURN(0); copy_out: /* Record is placed in temporary llog as is */ - rc = llog_write(env, mrul->temp_llh, rec, LLOG_NEXT_IDX); + rc = llog_write(env, mrd->temp_llh, rec, LLOG_NEXT_IDX); CDEBUG(D_MGS, "Copied idx=%d, rc=%d, len=%d, cmd %x %s %s\n", rec->lrh_index, rc, rec->lrh_len, lcfg->lcfg_command, @@ -1215,21 +1322,21 @@ static int mgs_log_is_empty(const struct lu_env *env, return rc; } -static int mgs_replace_nids_log(const struct lu_env *env, - struct obd_device *mgs, struct fs_db *fsdb, - char *logname, char *devname, char *nids) +static int mgs_replace_log(const struct lu_env *env, + struct obd_device *mgs, + char *logname, char *devname, + llog_cb_t replace_handler, void *data) { struct llog_handle *orig_llh, *backup_llh; struct llog_ctxt *ctxt; - struct mgs_replace_uuid_lookup *mrul; + struct mgs_replace_data *mrd; struct mgs_device *mgs_dev = lu2mgs_dev(mgs->obd_lu_dev); static struct obd_uuid cfg_uuid = { .uuid = "config_uuid" }; char *backup; - int rc, rc2; + int rc, rc2, buf_size; + time64_t now; ENTRY; - CDEBUG(D_MGS, "Replace nids for %s in %s\n", devname, logname); - ctxt = llog_get_context(mgs, LLOG_CONFIG_ORIG_CTXT); LASSERT(ctxt != NULL); @@ -1238,11 +1345,15 @@ static int mgs_replace_nids_log(const struct lu_env *env, GOTO(out_put, rc = 0); } - OBD_ALLOC(backup, strlen(logname) + strlen(".bak") + 1); + now = ktime_get_real_seconds(); + + /* max time64_t in decimal fits into 20 bytes long string */ + buf_size = strlen(logname) + 1 + 20 + 1 + strlen(".bak") + 1; + OBD_ALLOC(backup, buf_size); if (backup == NULL) GOTO(out_put, rc = -ENOMEM); - sprintf(backup, "%s.bak", logname); + snprintf(backup, buf_size, "%s.%llu.bak", logname, now); rc = llog_backup(env, mgs, ctxt, ctxt, logname, backup); if (rc == 0) { @@ -1279,20 +1390,23 @@ static int mgs_replace_nids_log(const struct lu_env *env, if (llog_get_size(backup_llh) <= 1) GOTO(out_close, rc = 0); - OBD_ALLOC_PTR(mrul); - if (!mrul) + OBD_ALLOC_PTR(mrd); + if (!mrd) GOTO(out_close, rc = -ENOMEM); /* devname is only needed information to replace UUID records */ - strlcpy(mrul->target.mti_svname, devname, - sizeof(mrul->target.mti_svname)); - /* parse nids later */ - strlcpy(mrul->target.mti_params, nids, sizeof(mrul->target.mti_params)); + if (devname) + strlcpy(mrd->target.mti_svname, devname, + sizeof(mrd->target.mti_svname)); + /* data is parsed in llog callback */ + if (data) + strlcpy(mrd->target.mti_params, data, + sizeof(mrd->target.mti_params)); /* Copy records to this temporary llog */ - mrul->temp_llh = orig_llh; + mrd->temp_llh = orig_llh; - rc = llog_process(env, backup_llh, mgs_replace_handler, - (void *)mrul, NULL); - OBD_FREE_PTR(mrul); + rc = llog_process(env, backup_llh, replace_handler, + (void *)mrd, NULL); + OBD_FREE_PTR(mrd); out_close: rc2 = llog_close(NULL, backup_llh); if (!rc) @@ -1314,30 +1428,55 @@ out_restore: } out_free: - OBD_FREE(backup, strlen(backup) + 1); + OBD_FREE(backup, buf_size); out_put: llog_ctxt_put(ctxt); if (rc) - CERROR("%s: failed to replace nids in log %s: rc = %d\n", + CERROR("%s: failed to replace log %s: rc = %d\n", mgs->obd_name, logname, rc); RETURN(rc); } +static int mgs_replace_nids_log(const struct lu_env *env, + struct obd_device *obd, + char *logname, char *devname, char *nids) +{ + CDEBUG(D_MGS, "Replace NIDs for %s in %s\n", devname, logname); + return mgs_replace_log(env, obd, logname, devname, + mgs_replace_nids_handler, nids); +} + /** * Parse device name and get file system name and/or device index * - * \param[in] devname device name (ex. lustre-MDT0000) - * \param[out] fsname file system name(optional) - * \param[out] index device index(optional) + * @devname device name (ex. lustre-MDT0000) + * @fsname file system name extracted from @devname and returned + * to the caller (optional) + * @index device index extracted from @devname and returned to + * the caller (optional) * - * \retval 0 success + * RETURN 0 success if we are only interested in + * extracting fsname from devname. + * i.e index is NULL + * + * LDD_F_SV_TYPE_* Besides extracting the fsname the + * user also wants the index. Report to + * the user the type of obd device the + * returned index belongs too. + * + * -EINVAL The obd device name is improper so + * fsname could not be extracted. + * + * -ENXIO Failed to extract the index out of + * the obd device name. Most likely an + * invalid obd device name */ -static int mgs_parse_devname(char *devname, char *fsname, __u32 *index) +static int mgs_parse_devname(char *devname, char *fsname, u32 *index) { - int rc; + int rc = 0; ENTRY; /* Extract fsname */ @@ -1355,11 +1494,12 @@ static int mgs_parse_devname(char *devname, char *fsname, __u32 *index) if (rc < 0) { CDEBUG(D_MGS, "Device name %s with wrong index\n", devname); - RETURN(-EINVAL); + RETURN(-ENXIO); } } - RETURN(0); + /* server_name2index can return LDD_F_SV_TYPE_* so always return rc */ + RETURN(rc); } /* This is only called during replace_nids */ @@ -1394,12 +1534,15 @@ static int only_mgs_is_running(struct obd_device *mgs_obd) return (num_devices <= 3) && (num_exports == 0); } -static int name_create_mdt(char **logname, char *fsname, int i) +static int name_create_mdt(char **logname, char *fsname, int mdt_idx) { - char mdt_index[9]; + char postfix[9]; + + if (mdt_idx > INDEX_MAP_MAX_VALUE) + return -E2BIG; - sprintf(mdt_index, "-MDT%04x", i); - return name_create(logname, fsname, mdt_index); + snprintf(postfix, sizeof(postfix), "-MDT%04x", mdt_idx); + return name_create(logname, fsname, postfix); } /** @@ -1439,9 +1582,9 @@ int mgs_replace_nids(const struct lu_env *env, GOTO(out, rc = -EINPROGRESS); } - /* Get fsname and index*/ + /* Get fsname and index */ rc = mgs_parse_devname(devname, fsname, &index); - if (rc) + if (rc < 0) GOTO(out, rc); rc = mgs_find_or_make_fsdb(env, mgs, fsname, &fsdb); @@ -1451,8 +1594,10 @@ int mgs_replace_nids(const struct lu_env *env, } /* Process client llogs */ - name_create(&logname, fsname, "-client"); - rc = mgs_replace_nids_log(env, mgs_obd, fsdb, logname, devname, nids); + rc = name_create(&logname, fsname, "-client"); + if (rc) + GOTO(out, rc); + rc = mgs_replace_nids_log(env, mgs_obd, logname, devname, nids); name_destroy(&logname); if (rc) { CERROR("%s: error while replacing NIDs for %s: rc = %d\n", @@ -1464,8 +1609,10 @@ int mgs_replace_nids(const struct lu_env *env, for (i = 0; i < INDEX_MAP_SIZE * 8; i++) { if (!test_bit(i, fsdb->fsdb_mdt_index_map)) continue; - name_create_mdt(&logname, fsname, i); - rc = mgs_replace_nids_log(env, mgs_obd, fsdb, logname, devname, nids); + rc = name_create_mdt(&logname, fsname, i); + if (rc) + GOTO(out, rc); + rc = mgs_replace_nids_log(env, mgs_obd, logname, devname, nids); name_destroy(&logname); if (rc) GOTO(out, rc); @@ -1482,6 +1629,188 @@ out: RETURN(rc); } +/** + * This is called for every record in llog. Some of records are + * skipped, others are copied to new log as is. + * Records to be skipped are + * marker records marked SKIP + * records enclosed between SKIP markers + * + * \param[in] llh log to be processed + * \param[in] rec current record + * \param[in] data mgs_replace_data structure + * + * \retval 0 success + **/ +static int mgs_clear_config_handler(const struct lu_env *env, + struct llog_handle *llh, + struct llog_rec_hdr *rec, void *data) +{ + struct mgs_replace_data *mrd; + struct lustre_cfg *lcfg = REC_DATA(rec); + int cfg_len = REC_DATA_LEN(rec); + int rc; + + ENTRY; + + mrd = (struct mgs_replace_data *)data; + + if (rec->lrh_type != OBD_CFG_REC) { + CDEBUG(D_MGS, "Config llog Name=%s, Record Index=%u, " + "Unhandled Record Type=%#x\n", llh->lgh_name, + rec->lrh_index, rec->lrh_type); + RETURN(-EINVAL); + } + + rc = lustre_cfg_sanity_check(lcfg, cfg_len); + if (rc) { + CDEBUG(D_MGS, "Config llog Name=%s, Invalid config file.", + llh->lgh_name); + RETURN(-EINVAL); + } + + if (lcfg->lcfg_command == LCFG_MARKER) { + struct cfg_marker *marker; + + marker = lustre_cfg_buf(lcfg, 1); + if (marker->cm_flags & CM_SKIP) { + if (marker->cm_flags & CM_START) + mrd->state = REPLACE_SKIP; + if (marker->cm_flags & CM_END) + mrd->state = REPLACE_COPY; + /* SKIP section started or finished */ + CDEBUG(D_MGS, "Skip idx=%d, rc=%d, len=%d, " + "cmd %x %s %s\n", rec->lrh_index, rc, + rec->lrh_len, lcfg->lcfg_command, + lustre_cfg_string(lcfg, 0), + lustre_cfg_string(lcfg, 1)); + RETURN(0); + } + } else { + if (mrd->state == REPLACE_SKIP) { + /* record enclosed between SKIP markers, skip it */ + CDEBUG(D_MGS, "Skip idx=%d, rc=%d, len=%d, " + "cmd %x %s %s\n", rec->lrh_index, rc, + rec->lrh_len, lcfg->lcfg_command, + lustre_cfg_string(lcfg, 0), + lustre_cfg_string(lcfg, 1)); + RETURN(0); + } + } + + /* Record is placed in temporary llog as is */ + rc = llog_write(env, mrd->temp_llh, rec, LLOG_NEXT_IDX); + + CDEBUG(D_MGS, "Copied idx=%d, rc=%d, len=%d, cmd %x %s %s\n", + rec->lrh_index, rc, rec->lrh_len, lcfg->lcfg_command, + lustre_cfg_string(lcfg, 0), lustre_cfg_string(lcfg, 1)); + RETURN(rc); +} + +/* + * Directory CONFIGS/ may contain files which are not config logs to + * be cleared. Skip any llogs with a non-alphanumeric character after + * the last '-'. For example, fsname-MDT0000.sav, fsname-MDT0000.bak, + * fsname-MDT0000.orig, fsname-MDT0000~, fsname-MDT0000.20150516, etc. + */ +static bool config_to_clear(const char *logname) +{ + int i; + char *str; + + str = strrchr(logname, '-'); + if (!str) + return 0; + + i = 0; + while (isalnum(str[++i])); + return str[i] == '\0'; +} + +/** + * Clear config logs for \a name + * + * \param env + * \param mgs MGS device + * \param name name of device or of filesystem + * (ex. lustre-OST0000 or lustre) in later case all logs + * will be cleared + * + * \retval 0 success + */ +int mgs_clear_configs(const struct lu_env *env, + struct mgs_device *mgs, const char *name) +{ + struct list_head dentry_list; + struct mgs_direntry *dirent, *n; + char *namedash; + int conn_state; + struct obd_device *mgs_obd = mgs->mgs_obd; + int rc; + + ENTRY; + + /* Prevent clients and servers from connecting to mgs */ + spin_lock(&mgs_obd->obd_dev_lock); + conn_state = mgs_obd->obd_no_conn; + mgs_obd->obd_no_conn = 1; + spin_unlock(&mgs_obd->obd_dev_lock); + + /* + * config logs cannot be cleaned if anything other than + * MGS is started + */ + if (!only_mgs_is_running(mgs_obd)) { + CERROR("Only MGS is allowed to be started\n"); + GOTO(out, rc = -EBUSY); + } + + /* Find all the logs in the CONFIGS directory */ + rc = class_dentry_readdir(env, mgs, &dentry_list); + if (rc) { + CERROR("%s: cannot read config directory '%s': rc = %d\n", + mgs_obd->obd_name, MOUNT_CONFIGS_DIR, rc); + GOTO(out, rc); + } + + if (list_empty(&dentry_list)) { + CERROR("%s: list empty reading config dir '%s': rc = %d\n", + mgs_obd->obd_name, MOUNT_CONFIGS_DIR, -ENOENT); + GOTO(out, rc = -ENOENT); + } + + OBD_ALLOC(namedash, strlen(name) + 2); + if (namedash == NULL) + GOTO(out, rc = -ENOMEM); + snprintf(namedash, strlen(name) + 2, "%s-", name); + + list_for_each_entry(dirent, &dentry_list, mde_list) { + if (strcmp(name, dirent->mde_name) && + strncmp(namedash, dirent->mde_name, strlen(namedash))) + continue; + if (!config_to_clear(dirent->mde_name)) + continue; + CDEBUG(D_MGS, "%s: Clear config log %s\n", + mgs_obd->obd_name, dirent->mde_name); + rc = mgs_replace_log(env, mgs_obd, dirent->mde_name, NULL, + mgs_clear_config_handler, NULL); + if (rc) + break; + } + + list_for_each_entry_safe(dirent, n, &dentry_list, mde_list) { + list_del_init(&dirent->mde_list); + mgs_direntry_free(dirent); + } + OBD_FREE(namedash, strlen(name) + 2); +out: + spin_lock(&mgs_obd->obd_dev_lock); + mgs_obd->obd_no_conn = conn_state; + spin_unlock(&mgs_obd->obd_dev_lock); + + RETURN(rc); +} + static int record_lov_setup(const struct lu_env *env, struct llog_handle *llh, char *devname, struct lov_desc *desc) { @@ -1569,7 +1898,7 @@ static int record_marker(const struct lu_env *env, sizeof(mgi->mgi_marker.cm_comment)); if (cplen >= sizeof(mgi->mgi_marker.cm_comment)) return -E2BIG; - mgi->mgi_marker.cm_createtime = cfs_time_current_sec(); + mgi->mgi_marker.cm_createtime = ktime_get_real_seconds(); mgi->mgi_marker.cm_canceltime = 0; lustre_cfg_bufs_reset(&mgi->mgi_bufs, NULL); lustre_cfg_bufs_set(&mgi->mgi_bufs, 1, &mgi->mgi_marker, @@ -1888,7 +2217,8 @@ static int mgs_steal_client_llog_handler(const struct lu_env *env, if (lcfg->lcfg_command == LCFG_SPTLRPC_CONF) RETURN(rc); - if (lcfg->lcfg_command == LCFG_ADD_MDC) { + if (lcfg->lcfg_command == LCFG_ADD_MDC && + strstr(lustre_cfg_string(lcfg, 0), "-clilmv") != NULL) { int index; if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1) @@ -2399,13 +2729,14 @@ static int mgs_write_log_mdt0(const struct lu_env *env, { char *log = mti->mti_svname; struct llog_handle *llh = NULL; - char *uuid, *lovname; + struct obd_uuid *uuid; + char *lovname; char mdt_index[6]; char *ptr = mti->mti_params; int rc = 0, failout = 0; ENTRY; - OBD_ALLOC(uuid, sizeof(struct obd_uuid)); + OBD_ALLOC_PTR(uuid); if (uuid == NULL) RETURN(-ENOMEM); @@ -2430,17 +2761,17 @@ static int mgs_write_log_mdt0(const struct lu_env *env, /* add MDT itself */ /* FIXME this whole fn should be a single journal transaction */ - sprintf(uuid, "%s_UUID", log); + sprintf(uuid->uuid, "%s_UUID", log); rc = record_marker(env, llh, fsdb, CM_START, log, "add mdt"); if (rc) GOTO(out_lod, rc); - rc = record_attach(env, llh, log, LUSTRE_MDT_NAME, uuid); + rc = record_attach(env, llh, log, LUSTRE_MDT_NAME, uuid->uuid); if (rc) GOTO(out_end, rc); rc = record_mount_opt(env, llh, log, lovname, NULL); if (rc) GOTO(out_end, rc); - rc = record_setup(env, llh, log, uuid, mdt_index, lovname, + rc = record_setup(env, llh, log, uuid->uuid, mdt_index, lovname, failout ? "n" : "f"); if (rc) GOTO(out_end, rc); @@ -2452,7 +2783,7 @@ out_end: out_lod: name_destroy(&lovname); out_free: - OBD_FREE(uuid, sizeof(struct obd_uuid)); + OBD_FREE_PTR(uuid); RETURN(rc); } @@ -2524,17 +2855,14 @@ static int mgs_write_log_mdt(const struct lu_env *env, if (rc) GOTO(out_free, rc); - rc = record_marker(env, llh, fsdb, CM_START, cliname, - "mount opts"); + rc = record_marker(env, llh, fsdb, CM_START, cliname, "mount opts"); if (rc) GOTO(out_end, rc); rc = record_mount_opt(env, llh, cliname, fsdb->fsdb_clilov, fsdb->fsdb_clilmv); if (rc) GOTO(out_end, rc); - rc = record_marker(env, llh, fsdb, CM_END, cliname, - "mount opts"); - + rc = record_marker(env, llh, fsdb, CM_END, cliname, "mount opts"); if (rc) GOTO(out_end, rc); @@ -2784,13 +3112,14 @@ out_end: /* Add ost to all MDT lov defs */ for (i = 0; i < INDEX_MAP_SIZE * 8; i++){ if (test_bit(i, fsdb->fsdb_mdt_index_map)) { - char mdt_index[9]; + char mdt_index[13]; rc = name_create_mdt_and_lov(&logname, &lovname, fsdb, i); if (rc) RETURN(rc); - sprintf(mdt_index, "-MDT%04x", i); + + snprintf(mdt_index, sizeof(mdt_index), "-MDT%04x", i); rc = mgs_write_log_osc_to_lov(env, mgs, fsdb, mti, logname, mdt_index, lovname, LUSTRE_SP_MDT, @@ -2884,7 +3213,8 @@ static int mgs_write_log_add_failnid(const struct lu_env *env, /* FIXME we currently can't erase the failnids * given when a target first registers, since they aren't part of - * an "add uuid" stanza */ + * an "add uuid" stanza + */ /* Verify that we know about this target */ if (mgs_log_is_empty(env, mgs, mti->mti_svname)) { @@ -2904,12 +3234,14 @@ static int mgs_write_log_add_failnid(const struct lu_env *env, } if (rc) RETURN(rc); + /* Add failover nids to the client log */ rc = name_create(&logname, mti->mti_fsname, "-client"); if (rc) { name_destroy(&cliname); RETURN(rc); } + rc = mgs_write_log_failnid_internal(env, mgs, fsdb,mti,logname,cliname); name_destroy(&logname); name_destroy(&cliname); @@ -2957,8 +3289,7 @@ static int mgs_wlp_lcfg(const struct lu_env *env, int rc, del; /* Erase any old settings of this same parameter */ - memcpy(comment, ptr, MTI_NAME_MAXLEN); - comment[MTI_NAME_MAXLEN - 1] = 0; + strlcpy(comment, ptr, sizeof(comment)); /* But don't try to match the value. */ tmp = strchr(comment, '='); if (tmp != NULL) @@ -2994,22 +3325,6 @@ static int mgs_wlp_lcfg(const struct lu_env *env, return rc; } -static int mgs_write_log_param2(const struct lu_env *env, - struct mgs_device *mgs, - struct fs_db *fsdb, - struct mgs_target_info *mti, char *ptr) -{ - struct lustre_cfg_bufs bufs; - int rc = 0; - ENTRY; - - CDEBUG(D_MGS, "next param '%s'\n", ptr); - rc = mgs_wlp_lcfg(env, mgs, fsdb, mti, PARAMS_FILENAME, &bufs, - mti->mti_svname, ptr); - - RETURN(rc); -} - /* write global variable settings into log */ static int mgs_write_log_sys(const struct lu_env *env, struct mgs_device *mgs, struct fs_db *fsdb, @@ -3053,7 +3368,14 @@ static int mgs_write_log_sys(const struct lu_env *env, return -ENOMEM; lcfg = &lcr->lcr_cfg; - lcfg->lcfg_num = convert ? simple_strtoul(tmp, NULL, 0) : 0; + if (convert) { + rc = kstrtouint(tmp, 0, &lcfg->lcfg_num); + if (rc) + GOTO(out_rec_free, rc); + } else { + lcfg->lcfg_num = 0; + } + /* truncate the comment to the parameter name */ ptr = tmp - 1; sep = *ptr; @@ -3077,6 +3399,7 @@ static int mgs_write_log_sys(const struct lu_env *env, } } *ptr = sep; +out_rec_free: lustre_cfg_rec_free(lcr); return rc; } @@ -3106,6 +3429,7 @@ static int mgs_write_log_quota(const struct lu_env *env, struct mgs_device *mgs, CDEBUG(D_MGS, "global '%s'\n", quota); if (strchr(tmp, 'u') == NULL && strchr(tmp, 'g') == NULL && + strchr(tmp, 'p') == NULL && strcmp(tmp, "none") != 0) { CERROR("enable option(%s) isn't supported\n", tmp); return -EINVAL; @@ -3489,6 +3813,79 @@ out: RETURN(rc); } +static int mgs_write_log_param2(const struct lu_env *env, + struct mgs_device *mgs, + struct fs_db *fsdb, + struct mgs_target_info *mti, char *ptr) +{ + struct lustre_cfg_bufs bufs; + int rc; + + ENTRY; + CDEBUG(D_MGS, "next param '%s'\n", ptr); + + /* PARAM_MGSNODE and PARAM_NETWORK are set only when formating + * or during the inital mount. It can never change after that. + */ + if (!class_match_param(ptr, PARAM_MGSNODE, NULL) || + !class_match_param(ptr, PARAM_NETWORK, NULL)) { + rc = 0; + goto end; + } + + /* Processed in mgs_write_log_ost. Another value that can't + * be changed by lctl set_param -P. + */ + if (!class_match_param(ptr, PARAM_FAILMODE, NULL)) { + LCONSOLE_ERROR_MSG(0x169, + "%s can only be changed with tunefs.lustre and --writeconf\n", + ptr); + rc = -EPERM; + goto end; + } + + /* FIXME !!! Support for sptlrpc is incomplete. Currently the change + * doesn't transmit to the client. See LU-7183. + */ + if (!class_match_param(ptr, PARAM_SRPC, NULL)) { + rc = mgs_srpc_set_param(env, mgs, fsdb, mti, ptr); + goto end; + } + + /* Can't use class_match_param since ptr doesn't start with + * PARAM_FAILNODE. So we look for PARAM_FAILNODE contained in ptr. + */ + if (strstr(ptr, PARAM_FAILNODE)) { + /* Add a failover nidlist. We already processed failovers + * params for new targets in mgs_write_log_target. + */ + const char *param; + + /* can't use wildcards with failover.node */ + if (strchr(ptr, '*')) { + rc = -ENODEV; + goto end; + } + + param = strstr(ptr, PARAM_FAILNODE); + if (strlcpy(mti->mti_params, param, sizeof(mti->mti_params)) >= + sizeof(mti->mti_params)) { + rc = -E2BIG; + goto end; + } + + CDEBUG(D_MGS, "Adding failnode with param %s\n", + mti->mti_params); + rc = mgs_write_log_add_failnid(env, mgs, fsdb, mti); + goto end; + } + + rc = mgs_wlp_lcfg(env, mgs, fsdb, mti, PARAMS_FILENAME, &bufs, + mti->mti_svname, ptr); +end: + RETURN(rc); +} + /* Permanent settings of all parameters by writing into the appropriate * configuration logs. * A parameter with null value ("='\0'") means to erase it out of @@ -3519,16 +3916,16 @@ static int mgs_write_log_param(const struct lu_env *env, if (class_match_param(ptr, PARAM_NETWORK, NULL) == 0) GOTO(end, rc); - /* Processed in mgs_write_log_ost */ - if (class_match_param(ptr, PARAM_FAILMODE, NULL) == 0) { - if (mti->mti_flags & LDD_F_PARAM) { - LCONSOLE_ERROR_MSG(0x169, "%s can only be " - "changed with tunefs.lustre" - "and --writeconf\n", ptr); - rc = -EPERM; - } - GOTO(end, rc); - } + /* Processed in mgs_write_log_ost */ + if (class_match_param(ptr, PARAM_FAILMODE, NULL) == 0) { + if (mti->mti_flags & LDD_F_PARAM) { + LCONSOLE_ERROR_MSG(0x169, + "%s can only be changed with tunefs.lustre and --writeconf\n", + ptr); + rc = -EPERM; + } + GOTO(end, rc); + } if (class_match_param(ptr, PARAM_SRPC, NULL) == 0) { rc = mgs_srpc_set_param(env, mgs, fsdb, mti, ptr); @@ -3576,7 +3973,7 @@ static int mgs_write_log_param(const struct lu_env *env, LCONSOLE_ERROR_MSG(0x144, "%s: MDC0 can not be" " (de)activated.\n", mti->mti_svname); - GOTO(end, rc = -EINVAL); + GOTO(end, rc = -EPERM); } } @@ -3611,18 +4008,12 @@ static int mgs_write_log_param(const struct lu_env *env, } active_err: if (rc < 0) { - LCONSOLE_ERROR_MSG(0x145, "Couldn't find %s in" - "log (%d). No permanent " - "changes were made to the " - "config log.\n", + LCONSOLE_ERROR_MSG(0x145, + "Couldn't find %s in log (%d). No permanent changes were made to the config log.\n", mti->mti_svname, rc); if (test_bit(FSDB_OLDLOG14, &fsdb->fsdb_flags)) - LCONSOLE_ERROR_MSG(0x146, "This may be" - " because the log" - "is in the old 1.4" - "style. Consider " - " --writeconf to " - "update the logs.\n"); + LCONSOLE_ERROR_MSG(0x146, + "This may be because the log is in the old 1.4 style. Consider --writeconf to update the logs.\n"); GOTO(end, rc); } /* Fall through to osc/mdc proc for deactivating live @@ -3904,6 +4295,7 @@ active_err: /* All mdd., ost. and osd. params in proc */ if ((class_match_param(ptr, PARAM_MDD, NULL) == 0) || + (class_match_param(ptr, PARAM_LOD, NULL) == 0) || (class_match_param(ptr, PARAM_OST, NULL) == 0) || (class_match_param(ptr, PARAM_OSD, NULL) == 0)) { CDEBUG(D_MGS, "%.3s param %s\n", ptr, ptr + 4); @@ -3954,27 +4346,26 @@ int mgs_write_log_target(const struct lu_env *env, struct mgs_device *mgs, mutex_lock(&fsdb->fsdb_mutex); - if (mti->mti_flags & - (LDD_F_VIRGIN | LDD_F_UPGRADE14 | LDD_F_WRITECONF)) { - /* Generate a log from scratch */ - if (mti->mti_flags & LDD_F_SV_TYPE_MDT) { + if (mti->mti_flags & (LDD_F_VIRGIN | LDD_F_WRITECONF)) { + /* Generate a log from scratch */ + if (mti->mti_flags & LDD_F_SV_TYPE_MDT) { rc = mgs_write_log_mdt(env, mgs, fsdb, mti); - } else if (mti->mti_flags & LDD_F_SV_TYPE_OST) { + } else if (mti->mti_flags & LDD_F_SV_TYPE_OST) { rc = mgs_write_log_ost(env, mgs, fsdb, mti); - } else { - CERROR("Unknown target type %#x, can't create log for " - "%s\n", mti->mti_flags, mti->mti_svname); - } - if (rc) { - CERROR("Can't write logs for %s (%d)\n", - mti->mti_svname, rc); - GOTO(out_up, rc); - } - } else { - /* Just update the params from tunefs in mgs_write_log_params */ - CDEBUG(D_MGS, "Update params for %s\n", mti->mti_svname); - mti->mti_flags |= LDD_F_PARAM; - } + } else { + CERROR("Unknown target type %#x, can't create log for %s\n", + mti->mti_flags, mti->mti_svname); + } + if (rc) { + CERROR("Can't write logs for %s (%d)\n", + mti->mti_svname, rc); + GOTO(out_up, rc); + } + } else { + /* Just update the params from tunefs in mgs_write_log_params */ + CDEBUG(D_MGS, "Update params for %s\n", mti->mti_svname); + mti->mti_flags |= LDD_F_PARAM; + } /* allocate temporary buffer, where class_get_next_param will make copy of a current parameter */ @@ -4107,8 +4498,8 @@ int mgs_list_logs(const struct lu_env *env, struct mgs_device *mgs, list_del_init(&dirent->mde_list); suffix = strrchr(dirent->mde_name, '-'); if (suffix != NULL) { - l = snprintf(out, remains, "config log: $%s\n", - dirent->mde_name); + l = scnprintf(out, remains, "config_log: %s\n", + dirent->mde_name); out += l; remains -= l; } @@ -4311,6 +4702,7 @@ static int mgs_lcfg_fork_handler(const struct lu_env *env, } /* else case fall through */ } /* else case fall through */ } + /* fallthrough */ default: { for (i = 1; i < cnt; i++) { o_buflen = o_lcfg->lcfg_buflens[i]; @@ -4442,6 +4834,10 @@ int mgs_lcfg_fork(const struct lu_env *env, struct mgs_device *mgs, int rc = 0; ENTRY; + if (unlikely(!oldname || oldname[0] == '\0' || + !newname || newname[0] == '\0')) + RETURN(-EINVAL); + if (strcmp(oldname, newname) == 0) RETURN(-EINVAL); @@ -4512,33 +4908,147 @@ int mgs_lcfg_erase(const struct lu_env *env, struct mgs_device *mgs, int rc; ENTRY; + if (unlikely(!fsname || fsname[0] == '\0')) + RETURN(-EINVAL); + rc = mgs_erase_logs(env, mgs, fsname); RETURN(rc); } -/* from llog_swab */ -static void print_lustre_cfg(struct lustre_cfg *lcfg) +static int mgs_xattr_del(const struct lu_env *env, struct dt_object *obj) { - int i; - ENTRY; + struct dt_device *dev; + struct thandle *th = NULL; + int rc = 0; - CDEBUG(D_MGS, "lustre_cfg: %p\n", lcfg); - CDEBUG(D_MGS, "\tlcfg->lcfg_version: %#x\n", lcfg->lcfg_version); + ENTRY; - CDEBUG(D_MGS, "\tlcfg->lcfg_command: %#x\n", lcfg->lcfg_command); - CDEBUG(D_MGS, "\tlcfg->lcfg_num: %#x\n", lcfg->lcfg_num); - CDEBUG(D_MGS, "\tlcfg->lcfg_flags: %#x\n", lcfg->lcfg_flags); - CDEBUG(D_MGS, "\tlcfg->lcfg_nid: %s\n", libcfs_nid2str(lcfg->lcfg_nid)); + dev = container_of0(obj->do_lu.lo_dev, struct dt_device, dd_lu_dev); + th = dt_trans_create(env, dev); + if (IS_ERR(th)) + RETURN(PTR_ERR(th)); - CDEBUG(D_MGS, "\tlcfg->lcfg_bufcount: %d\n", lcfg->lcfg_bufcount); - if (lcfg->lcfg_bufcount < LUSTRE_CFG_MAX_BUFCOUNT) - for (i = 0; i < lcfg->lcfg_bufcount; i++) { - CDEBUG(D_MGS, "\tlcfg->lcfg_buflens[%d]: %d %s\n", - i, lcfg->lcfg_buflens[i], - lustre_cfg_string(lcfg, i)); - } - EXIT; + rc = dt_declare_xattr_del(env, obj, XATTR_TARGET_RENAME, th); + if (rc) + GOTO(stop, rc); + + rc = dt_trans_start_local(env, dev, th); + if (rc) + GOTO(stop, rc); + + dt_write_lock(env, obj, 0); + rc = dt_xattr_del(env, obj, XATTR_TARGET_RENAME, th); + + GOTO(unlock, rc); + +unlock: + dt_write_unlock(env, obj); + +stop: + dt_trans_stop(env, dev, th); + + return rc; +} + +int mgs_lcfg_rename(const struct lu_env *env, struct mgs_device *mgs) +{ + struct list_head log_list; + struct mgs_direntry *dirent, *n; + char fsname[16]; + struct lu_buf buf = { + .lb_buf = fsname, + .lb_len = sizeof(fsname) + }; + int rc = 0; + + ENTRY; + + rc = class_dentry_readdir(env, mgs, &log_list); + if (rc) + RETURN(rc); + + if (list_empty(&log_list)) + RETURN(0); + + list_for_each_entry_safe(dirent, n, &log_list, mde_list) { + struct dt_object *o = NULL; + char oldname[16]; + char *ptr; + int len; + + list_del_init(&dirent->mde_list); + ptr = strrchr(dirent->mde_name, '-'); + if (!ptr) + goto next; + + len = ptr - dirent->mde_name; + if (unlikely(len >= sizeof(oldname))) { + CDEBUG(D_MGS, "Skip invalid configuration file %s\n", + dirent->mde_name); + goto next; + } + + o = local_file_find(env, mgs->mgs_los, mgs->mgs_configs_dir, + dirent->mde_name); + if (IS_ERR(o)) { + rc = PTR_ERR(o); + CDEBUG(D_MGS, "Fail to locate file %s: rc = %d\n", + dirent->mde_name, rc); + goto next; + } + + rc = dt_xattr_get(env, o, &buf, XATTR_TARGET_RENAME); + if (rc < 0) { + if (rc == -ENODATA) + rc = 0; + else + CDEBUG(D_MGS, + "Fail to get EA for %s: rc = %d\n", + dirent->mde_name, rc); + goto next; + } + + if (unlikely(rc == len && + memcmp(fsname, dirent->mde_name, len) == 0)) { + /* The new fsname is the same as the old one. */ + rc = mgs_xattr_del(env, o); + goto next; + } + + memcpy(oldname, dirent->mde_name, len); + oldname[len] = '\0'; + fsname[rc] = '\0'; + rc = mgs_lcfg_fork_one(env, mgs, dirent, oldname, fsname); + if (rc && rc != -EEXIST) { + CDEBUG(D_MGS, "Fail to fork %s: rc = %d\n", + dirent->mde_name, rc); + goto next; + } + + rc = mgs_erase_log(env, mgs, dirent->mde_name); + if (rc) { + CDEBUG(D_MGS, "Fail to erase old %s: rc = %d\n", + dirent->mde_name, rc); + /* keep it there if failed to remove it. */ + rc = 0; + } + +next: + if (o && !IS_ERR(o)) + lu_object_put(env, &o->do_lu); + + mgs_direntry_free(dirent); + if (rc) + break; + } + + list_for_each_entry_safe(dirent, n, &log_list, mde_list) { + list_del_init(&dirent->mde_list); + mgs_direntry_free(dirent); + } + + RETURN(rc); } /* Setup _mgs fsdb and log @@ -4588,123 +5098,273 @@ int mgs_params_fsdb_cleanup(const struct lu_env *env, struct mgs_device *mgs) return rc == -ENOENT ? 0 : rc; } -/* Set a permanent (config log) param for a target or fs - * \param lcfg buf0 may contain the device (testfs-MDT0000) name - * buf1 contains the single parameter - */ -int mgs_setparam(const struct lu_env *env, struct mgs_device *mgs, - struct lustre_cfg *lcfg, char *fsname) +/** + * Fill in the mgs_target_info based on data devname and param provide. + * + * @env thread context + * @mgs mgs device + * @mti mgs target info. We want to set this based other paramters + * passed to this function. Once setup we write it to the config + * logs. + * @devname optional OBD device name + * @param string that contains both what tunable to set and the value to + * set it to. + * + * RETURN 0 for success + * negative error number on failure + **/ +static int mgs_set_conf_param(const struct lu_env *env, struct mgs_device *mgs, + struct mgs_target_info *mti, const char *devname, + const char *param) { struct fs_db *fsdb = NULL; - struct mgs_target_info *mti = NULL; - char *devname, *param; - char *ptr; - const char *tmp; - __u32 index; + int dev_type; int rc = 0; - bool free = false; + ENTRY; + /* lustre, lustre-mdtlov, lustre-client, lustre-MDT0000 */ + if (!devname) { + size_t len; + + /* We have two possible cases here: + * + * 1) the device name embedded in the param: + * lustre-OST0000.osc.max_dirty_mb=32 + * + * 2) the file system name is embedded in + * the param: lustre.sys.at.min=0 + */ + len = strcspn(param, ".="); + if (!len || param[len] == '=') + RETURN(-EINVAL); - print_lustre_cfg(lcfg); + if (len >= sizeof(mti->mti_svname)) + RETURN(-E2BIG); - /* lustre, lustre-mdtlov, lustre-client, lustre-MDT0000 */ - devname = lustre_cfg_string(lcfg, 0); - param = lustre_cfg_string(lcfg, 1); - if (!devname) { - /* Assume device name embedded in param: - lustre-OST0000.osc.max_dirty_mb=32 */ - ptr = strchr(param, '.'); - if (ptr) { - devname = param; - *ptr = 0; - param = ptr + 1; - } - } - if (!devname) { - LCONSOLE_ERROR_MSG(0x14d, "No target specified: %s\n", param); - RETURN(-ENOSYS); - } + snprintf(mti->mti_svname, sizeof(mti->mti_svname), + "%.*s", (int)len, param); + param += len + 1; + } else { + if (strlcpy(mti->mti_svname, devname, sizeof(mti->mti_svname)) >= + sizeof(mti->mti_svname)) + RETURN(-E2BIG); + } - rc = mgs_parse_devname(devname, fsname, NULL); - if (rc == 0 && !mgs_parse_devname(devname, NULL, &index)) { - /* param related to llite isn't allowed to set by OST or MDT */ - if (rc == 0 && strncmp(param, PARAM_LLITE, - sizeof(PARAM_LLITE) - 1) == 0) - RETURN(-EINVAL); - } else { - /* assume devname is the fsname */ - strlcpy(fsname, devname, MTI_NAME_MAXLEN); - } - CDEBUG(D_MGS, "setparam fs='%s' device='%s'\n", fsname, devname); + if (!strlen(mti->mti_svname)) { + LCONSOLE_ERROR_MSG(0x14d, "No target specified: %s\n", param); + RETURN(-ENOSYS); + } - rc = mgs_find_or_make_fsdb(env, mgs, - lcfg->lcfg_command == LCFG_SET_PARAM ? - PARAMS_FILENAME : fsname, &fsdb); + dev_type = mgs_parse_devname(mti->mti_svname, mti->mti_fsname, + &mti->mti_stripe_index); + switch (dev_type) { + /* For this case we have an invalid obd device name */ + case -ENXIO: + CDEBUG(D_MGS, "%s don't contain an index\n", mti->mti_svname); + strlcpy(mti->mti_fsname, mti->mti_svname, MTI_NAME_MAXLEN); + dev_type = 0; + break; + /* Not an obd device, assume devname is the fsname. + * User might of only provided fsname and not obd device + */ + case -EINVAL: + CDEBUG(D_MGS, "%s is seen as a file system name\n", mti->mti_svname); + strlcpy(mti->mti_fsname, mti->mti_svname, MTI_NAME_MAXLEN); + dev_type = 0; + break; + default: + if (dev_type < 0) + GOTO(out, rc = dev_type); + + /* param related to llite isn't allowed to set by OST or MDT */ + if (dev_type & LDD_F_SV_TYPE_OST || + dev_type & LDD_F_SV_TYPE_MDT) { + /* param related to llite isn't allowed to set by OST + * or MDT + */ + if (!strncmp(param, PARAM_LLITE, + sizeof(PARAM_LLITE) - 1)) + GOTO(out, rc = -EINVAL); + + /* Strip -osc or -mdc suffix from svname */ + if (server_make_name(dev_type, mti->mti_stripe_index, + mti->mti_fsname, mti->mti_svname, + sizeof(mti->mti_svname))) + GOTO(out, rc = -EINVAL); + } + break; + } + + if (strlcpy(mti->mti_params, param, sizeof(mti->mti_params)) >= + sizeof(mti->mti_params)) + GOTO(out, rc = -E2BIG); + + CDEBUG(D_MGS, "set_conf_param fs='%s' device='%s' param='%s'\n", + mti->mti_fsname, mti->mti_svname, mti->mti_params); + + rc = mgs_find_or_make_fsdb(env, mgs, mti->mti_fsname, &fsdb); if (rc) - RETURN(rc); + GOTO(out, rc); - if (lcfg->lcfg_command != LCFG_SET_PARAM && - !test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags) && + if (!test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags) && test_bit(FSDB_LOG_EMPTY, &fsdb->fsdb_flags)) { CERROR("No filesystem targets for %s. cfg_device from lctl " - "is '%s'\n", fsname, devname); - free = true; + "is '%s'\n", mti->mti_fsname, mti->mti_svname); + mgs_unlink_fsdb(mgs, fsdb); GOTO(out, rc = -EINVAL); } - /* Create a fake mti to hold everything */ - OBD_ALLOC_PTR(mti); - if (!mti) - GOTO(out, rc = -ENOMEM); - if (strlcpy(mti->mti_fsname, fsname, sizeof(mti->mti_fsname)) - >= sizeof(mti->mti_fsname)) - GOTO(out, rc = -E2BIG); - if (strlcpy(mti->mti_svname, devname, sizeof(mti->mti_svname)) - >= sizeof(mti->mti_svname)) - GOTO(out, rc = -E2BIG); - if (strlcpy(mti->mti_params, param, sizeof(mti->mti_params)) - >= sizeof(mti->mti_params)) - GOTO(out, rc = -E2BIG); - rc = server_name2index(mti->mti_svname, &mti->mti_stripe_index, &tmp); - if (rc < 0) - /* Not a valid server; may be only fsname */ - rc = 0; - else - /* Strip -osc or -mdc suffix from svname */ - if (server_make_name(rc, mti->mti_stripe_index, mti->mti_fsname, - mti->mti_svname)) - GOTO(out, rc = -EINVAL); /* * Revoke lock so everyone updates. Should be alright if * someone was already reading while we were updating the logs, * so we don't really need to hold the lock while we're * writing (above). */ - if (lcfg->lcfg_command == LCFG_SET_PARAM) { - mti->mti_flags = rc | LDD_F_PARAM2; - mutex_lock(&fsdb->fsdb_mutex); - rc = mgs_write_log_param2(env, mgs, fsdb, mti, mti->mti_params); - mutex_unlock(&fsdb->fsdb_mutex); - mgs_revoke_lock(mgs, fsdb, CONFIG_T_PARAMS); + mti->mti_flags = dev_type | LDD_F_PARAM; + mutex_lock(&fsdb->fsdb_mutex); + rc = mgs_write_log_param(env, mgs, fsdb, mti, mti->mti_params); + mutex_unlock(&fsdb->fsdb_mutex); + mgs_revoke_lock(mgs, fsdb, CONFIG_T_CONFIG); + +out: + if (fsdb) + mgs_put_fsdb(mgs, fsdb); + + RETURN(rc); +} + +static int mgs_set_param2(const struct lu_env *env, struct mgs_device *mgs, + struct mgs_target_info *mti, const char *param) +{ + struct fs_db *fsdb = NULL; + int dev_type; + size_t len; + int rc; + + if (strlcpy(mti->mti_params, param, sizeof(mti->mti_params)) >= + sizeof(mti->mti_params)) + GOTO(out, rc = -E2BIG); + + len = strcspn(param, ".="); + if (len && param[len] != '=') { + struct list_head *tmp; + char *ptr; + + param += len + 1; + ptr = strchr(param, '.'); + + len = strlen(param); + if (ptr) + len -= strlen(ptr); + if (len >= sizeof(mti->mti_svname)) + GOTO(out, rc = -E2BIG); + + snprintf(mti->mti_svname, sizeof(mti->mti_svname), "%.*s", + (int)len, param); + + mutex_lock(&mgs->mgs_mutex); + if (unlikely(list_empty(&mgs->mgs_fs_db_list))) { + mutex_unlock(&mgs->mgs_mutex); + GOTO(out, rc = -ENODEV); + } + + list_for_each(tmp, &mgs->mgs_fs_db_list) { + fsdb = list_entry(tmp, struct fs_db, fsdb_list); + if (fsdb->fsdb_has_lproc_entry && + strcmp(fsdb->fsdb_name, "params") != 0 && + strstr(param, fsdb->fsdb_name)) { + snprintf(mti->mti_svname, + sizeof(mti->mti_svname), "%s", + fsdb->fsdb_name); + break; + } + fsdb = NULL; + } + + if (!fsdb) { + snprintf(mti->mti_svname, sizeof(mti->mti_svname), + "general"); + } + mutex_unlock(&mgs->mgs_mutex); } else { - mti->mti_flags = rc | LDD_F_PARAM; - mutex_lock(&fsdb->fsdb_mutex); - rc = mgs_write_log_param(env, mgs, fsdb, mti, mti->mti_params); - mutex_unlock(&fsdb->fsdb_mutex); - mgs_revoke_lock(mgs, fsdb, CONFIG_T_CONFIG); + snprintf(mti->mti_svname, sizeof(mti->mti_svname), "general"); } + CDEBUG(D_MGS, "set_param2 fs='%s' device='%s' param='%s'\n", + mti->mti_fsname, mti->mti_svname, mti->mti_params); + + /* The return value should be the device type i.e LDD_F_SV_TYPE_XXX. + * A returned error tells us we don't have a target obd device. + */ + dev_type = server_name2index(mti->mti_svname, &mti->mti_stripe_index, + NULL); + if (dev_type < 0) + dev_type = 0; + + /* the return value should be the device type i.e LDD_F_SV_TYPE_XXX. + * Strip -osc or -mdc suffix from svname + */ + if ((dev_type & LDD_F_SV_TYPE_OST || dev_type & LDD_F_SV_TYPE_MDT) && + server_make_name(dev_type, mti->mti_stripe_index, + mti->mti_fsname, mti->mti_svname, + sizeof(mti->mti_svname))) + GOTO(out, rc = -EINVAL); + + rc = mgs_find_or_make_fsdb(env, mgs, PARAMS_FILENAME, &fsdb); + if (rc) + GOTO(out, rc); + /* + * Revoke lock so everyone updates. Should be alright if + * someone was already reading while we were updating the logs, + * so we don't really need to hold the lock while we're + * writing (above). + */ + mti->mti_flags = dev_type | LDD_F_PARAM2; + mutex_lock(&fsdb->fsdb_mutex); + rc = mgs_write_log_param2(env, mgs, fsdb, mti, mti->mti_params); + mutex_unlock(&fsdb->fsdb_mutex); + mgs_revoke_lock(mgs, fsdb, CONFIG_T_PARAMS); + mgs_put_fsdb(mgs, fsdb); out: - if (mti) - OBD_FREE_PTR(mti); + RETURN(rc); +} - if (fsdb) { - if (free) - mgs_unlink_fsdb(mgs, fsdb); - mgs_put_fsdb(mgs, fsdb); +/* Set a permanent (config log) param for a target or fs + * + * @lcfg buf0 may contain the device (testfs-MDT0000) name + * buf1 contains the single parameter + */ +int mgs_set_param(const struct lu_env *env, struct mgs_device *mgs, + struct lustre_cfg *lcfg) +{ + const char *param = lustre_cfg_string(lcfg, 1); + struct mgs_target_info *mti; + int rc; + + /* Create a fake mti to hold everything */ + OBD_ALLOC_PTR(mti); + if (!mti) + return -ENOMEM; + + print_lustre_cfg(lcfg); + + if (lcfg->lcfg_command == LCFG_PARAM) { + /* For the case of lctl conf_param devname can be + * lustre, lustre-mdtlov, lustre-client, lustre-MDT0000 + */ + const char *devname = lustre_cfg_string(lcfg, 0); + + rc = mgs_set_conf_param(env, mgs, mti, devname, param); + } else { + /* In the case of lctl set_param -P lcfg[0] will always + * be 'general'. At least for now. + */ + rc = mgs_set_param2(env, mgs, mti, param); } - RETURN(rc); + OBD_FREE_PTR(mti); + + return rc; } static int mgs_write_log_pool(const struct lu_env *env, @@ -4737,13 +5397,13 @@ int mgs_nodemap_cmd(const struct lu_env *env, struct mgs_device *mgs, enum lcfg_command_type cmd, const char *nodemap_name, char *param) { - lnet_nid_t nid[2]; - __u32 idmap[2]; - bool bool_switch; - __u32 int_id; - int rc = 0; - ENTRY; + lnet_nid_t nid[2]; + u32 idmap[2]; + bool bool_switch; + u32 int_id; + int rc = 0; + ENTRY; switch (cmd) { case LCFG_NODEMAP_ADD: rc = nodemap_add(nodemap_name); @@ -4764,23 +5424,51 @@ int mgs_nodemap_cmd(const struct lu_env *env, struct mgs_device *mgs, rc = nodemap_del_range(nodemap_name, nid); break; case LCFG_NODEMAP_ADMIN: - bool_switch = simple_strtoul(param, NULL, 10); + rc = kstrtobool(param, &bool_switch); + if (rc) + break; rc = nodemap_set_allow_root(nodemap_name, bool_switch); break; case LCFG_NODEMAP_DENY_UNKNOWN: - bool_switch = simple_strtoul(param, NULL, 10); + rc = kstrtobool(param, &bool_switch); + if (rc) + break; rc = nodemap_set_deny_unknown(nodemap_name, bool_switch); break; + case LCFG_NODEMAP_AUDIT_MODE: + rc = kstrtoul(param, 10, (unsigned long *)&bool_switch); + if (rc == 0) + rc = nodemap_set_audit_mode(nodemap_name, bool_switch); + break; + case LCFG_NODEMAP_MAP_MODE: + if (strcmp("both", param) == 0) + rc = nodemap_set_mapping_mode(nodemap_name, + NODEMAP_MAP_BOTH); + else if (strcmp("uid_only", param) == 0) + rc = nodemap_set_mapping_mode(nodemap_name, + NODEMAP_MAP_UID_ONLY); + else if (strcmp("gid_only", param) == 0) + rc = nodemap_set_mapping_mode(nodemap_name, + NODEMAP_MAP_GID_ONLY); + else + rc = -EINVAL; + break; case LCFG_NODEMAP_TRUSTED: - bool_switch = simple_strtoul(param, NULL, 10); + rc = kstrtobool(param, &bool_switch); + if (rc) + break; rc = nodemap_set_trust_client_ids(nodemap_name, bool_switch); break; case LCFG_NODEMAP_SQUASH_UID: - int_id = simple_strtoul(param, NULL, 10); + rc = kstrtouint(param, 10, &int_id); + if (rc) + break; rc = nodemap_set_squash_uid(nodemap_name, int_id); break; case LCFG_NODEMAP_SQUASH_GID: - int_id = simple_strtoul(param, NULL, 10); + rc = kstrtouint(param, 10, &int_id); + if (rc) + break; rc = nodemap_set_squash_gid(nodemap_name, int_id); break; case LCFG_NODEMAP_ADD_UIDMAP: @@ -4810,6 +5498,9 @@ int mgs_nodemap_cmd(const struct lu_env *env, struct mgs_device *mgs, case LCFG_NODEMAP_SET_FILESET: rc = nodemap_set_fileset(nodemap_name, param); break; + case LCFG_NODEMAP_SET_SEPOL: + rc = nodemap_set_sepol(nodemap_name, param); + break; default: rc = -EINVAL; }